Thereβs only a handful of people I trust β and even fewer AI detection tools.
The reason behind this is simple. Almost every article I read about AI detection concerns a student getting hit with a false positive. This isnβt a good look, especially since we use these platforms to give us some clarity during this uncertain time of AI.
That said, if Iβm in need of a detector, the one I always use is Sapling. Compared to other tools, I find it to be accurate and straightforward. Oh, and did I mention that itβs also free?
Now, the question is, will it hold up in an in-depth review?
What is Sapling AI?
Looking for a free AI detection tool? Look no further than Sapling AI.
This is the brainchild of developers from UC Berkeley, Stanford, Google, and Meta β so you know itβs going to be good. Sapling uses transformer technology to identify phrases from a text that are likely to be AI and outputs a likelihood score based on certainty. At its current form, it claims to detect text from GPT, Claude, and Gemini with a high accuracy.
Does Sapling Have Any Features?
Not much.
Unlike other AI detection tools like Copyleaks and Winston AI, Saplingβs version of this tool is pretty barebones. It does output an AI likelihood score, allows .doc and .pdf uploads, and highlights parts of the text according to how likely they are to be from machines, but apart from those, Sapling doesnβt offer much in terms of features.
However, according to their roadmap, theyβre implementing a feature in the future that allows you to check if a piece of code is AI-generated, but weβll have to stay tuned for that.
Is Sapling Accurate?
According to Sapling, their detection is 97% accurate. According to, well, us, Sapling has an 87.04% accuracy rate in true positive tests and 93.84% in false positive tests β making it the most accurate detector out of all we tested along with Winston, Originality, and Copyleaks.
But are they consistent? Letβs find out by testing it again.
Test #1
Sapling: Successfully detects ChatGPT text as AI-generated.
AI Likelihood Score: 100%
Test #2
Sapling: Successfully detects ChatGPT text as AI-generated.
AI Likelihood Score: 100%
Test #3
Sapling: Successfully detects ChatGPT text as AI-generated.
AI Likelihood Score: 100%
Test #4
Sapling: Successfully detects ChatGPT text as AI-generated.
AI Likelihood Score: 100%
Test #5
Sapling: Successfully detects ChatGPT text as AI-generated.
AI Likelihood Score: 100%
Test #6
Sapling: Successfully detects ChatGPT text as AI-generated.
AI Likelihood Score: 100%
Test #7
Sapling: Successfully detects ChatGPT text as AI-generated.
AI Likelihood Score: 100%
Test #8
Sapling: Successfully detects ChatGPT text as AI-generated.
AI Likelihood Score: 100%
Overall Tally
|
Sapling vs. ChatGPT AI Likelihood Score |
|
Sapling AI vs. Undetectable AI
At the bottom of Saplingβs homepage is this note:
Interesting note β so I figured, why donβt I give it a crack?
Iβm going to pit Sapling against Undetectable AI: a popular AI bypassing platform that can tweak any text so that it could avoid AI detection. This works by removing common AI markers and cleverly placing intentional errors to trick detectors. Weβve reviewed this platform before and tested it against AI detectors and other paraphrasers, and it hasnβt let us down yet.Β
Will it falter against Sapling? Letβs find out.
Test #1
Sapling: Successfully detects ChatGPT text as AI-generated.
AI Likelihood Score: 100%
Test #2
Sapling: Sapling categorizes AI-generated text as human.
AI Likelihood Score: 0.1%
Test #3
Sapling: Sapling categorizes AI-generated text as human.
AI Likelihood Score: 0%
Test #4
Sapling: Sapling categorizes AI-generated text as human.
AI Likelihood Score: 0%
Test #5
Sapling: Sapling categorizes AI-generated text as human.
AI Likelihood Score: 0.3%
Test #6
Sapling: Sapling categorizes AI-generated text as human.
AI Likelihood Score: 31%
Test #7
Sapling: Successfully detects ChatGPT text as AI-generated.
AI Likelihood Score: 75.6%
Overall Tally
|
Sapling vs. Undetectable AI AI Likelihood Score |
|
Sapling AIβs Pros and Cons
|
|
The Final Verdict
Among the free AI detection tools in the market, itβs hard to find one as reliable as Sapling.
Sure, it doesnβt have many features β but what else do you need, really? It has an AI detection tool that works (for the most part). It has some pitfalls when measured against AI bypassers, but Undetectable AI is made to mess with detectors, so I canβt penalize Sapling too much.
Sapling is good enough to hang with its paid counterparts like Winston AI. And at the end of the day, determining whether a document is AI-generated shouldnβt end with detection tools anyway. We need to exercise our own judgment so that no innocent students will pay the price.
Looking for more information about AI detection? This article is a good starting point. Good luck!
